Manufacturing & auto components

Remaining useful life for rotating machinery

Sensor-driven prognostics that say how many cycles a machine has left, so maintenance is scheduled by condition rather than calendar.

Degradation curves descending toward a threshold line
Typical pilot
6–10 weeks on one asset class
Benchmark
NASA C-MAPSS Turbofan Degradation
Benchmark host
NASA Prognostics Center of Excellence
Sector
Manufacturing & auto components
What this is — and is not

This page describes a problem class and the architecture we deploy for it, written against a named public benchmark — the open dataset or competition where the world's data scientists have tested approaches to this exact problem against a hard metric. It is not a client engagement, and the benchmark's results, prizes and rankings belong to its host and participants, not to us. Client work is confidential and is only ever published with written permission.

The problem

What it costs when this goes unsolved.

Calendar-based maintenance replaces healthy parts and still lets unhealthy ones fail early. The expensive version of that mistake is an unplanned line stop; the quiet version is a spares budget sized for the worst case. Condition-based scheduling needs one number per asset — how long it has left — with enough confidence to plan around.

The data reality

Run-to-failure histories are the scarcest data in industry, because well-run plants do not let machines run to failure. What exists is multivariate sensor telemetry — temperatures, pressures, vibration, shaft speeds — across operating regimes that shift the baselines. The classic modelling trap is treating degradation as linear from day one, when real assets hold steady and then decline.

How we build it

The architecture, stage by stage.

01

Health index before prediction

Sensor channels are fused into a degradation signal per asset, validated against known failures and maintenance records before any forecasting is attempted.

02

Sequence models with piecewise targets

LSTM, GRU and temporal-convolution models trained with piecewise-linear remaining-life targets — flat while the asset is healthy, declining once degradation onsets — which is what makes the estimates usable early in life.

03

Uncertainty you can plan with

Predictions ship as intervals, not point estimates, so a planner can choose the confidence level that matches the cost of being wrong in each direction.

04

Transfer across the fleet

The pipeline generalises across assets of the same class — engines, turbines, pumps, compressors — with per-asset calibration rather than per-asset models.

Model families on this problem class: LSTM / GRU / TCN sequence models · Health-index regression · Survival analysis · Historian integration.

What you get

What an engagement hands over.

Everything below goes in the scope document before you sign it, with a fixed price or a rate with a ceiling — the same terms as every other engagement in the catalogue.

  • A remaining-useful-life model per asset class, with confidence intervals
  • A maintenance-scheduling policy tied to your downtime and spares costs
  • Integration with your historian or CMMS for continuous scoring
  • An honest assessment of whether your data can support prognostics yet — and what to instrument if it cannot

Provenance

The benchmark behind this page.

NASA C-MAPSS Turbofan Degradation, run by NASA Prognostics Center of Excellence, is the public proving ground for this problem class. The figures below are the host's, cited as context for how seriously this problem is tested in the open — they are not our results and we do not claim them.

Benchmark scale
Run-to-failure traces · 21 sensor channels · 4 operating regimes
Standing
The reference benchmark of the prognostics and health-management field
Reference
Benchmark page ↗

Next step

Thirty minutes on whether this fits your problem.

Bring the constraint — the regulator, the data boundary, the latency budget. If your data cannot support this build yet, the call will conclude with what to fix first, not with a proposal.